Proposal Contact PI's Surname Initials Application …MARSDEN FUND FULL RESEARCH PROPOSAL Standard Application Form A new generation of statistical models for spatial point process

MARSDEN FUND FULL RESEARCH PROPOSALStandard Application Form

A new generation of statistical models for spatial point process data1A. TITLE OF RESEARCH PROPOSAL

03-479 7772Email [email protected]

1B. IDENTIFICATION

Dr Tilman DaviesDepartment of Mathematics & StatisticsUniversity of OtagoPO Box 56DUNEDINDUNEDIN 9054NEW ZEALAND

Name (with Title)Full Address

Contact Principal Investigator

Phone

Associate Investigator(s)Name (with title) Institution Country

Professor MartinHazelton

NEW ZEALANDMassey University

Professor AdrianBaddeley

AUSTRALIACurtin University

A dataset comprising the spatial locations of events or items of interest is called a spatial pointpattern. Such data arise in myriad areas, as diverse as epidemiology, ecology, and archaeology.Typically, models for the distribution of such data incorporate two fundamental components, onerelating to persistent spatial (fixed) effects and the other reflecting (stochastic) interactions betweenpoints, such as a tendency to cluster together. For example, for a pattern of disease cases the firstcomponent would relate to the variation in underlying population density while the latter couldrepresent infectious transmission. A fundamental task, with critical practical consequences, is to separate these components in theanalysis of any given pattern. This is notoriously difficult because of a lack of identifiability: inmost current models the two competing representations of spatial variation yield identicalpredictions, and therefore cannot be distinguished even with unlimited data. However, leveragingrecent developments on some specific examples, allied to progress in related areas of statistics, wewill develop a new, comprehensive suite of flexible 'hybrid' models incorporating both componentsin an identifiable manner, overcoming a decades-long stalemate on progress. This will allow us toattack important research questions that have previously remained elusive.

1D. SUMMARY

ProposalStandard

Contact PI 's SurnameDavies

InitialsTM

Application Number19-UOO-191

PanelMIS

2019 Marsden Fund Proposals - Full Round Page 1

2A. BACKGROUNDA spatial point pattern is a dataset recording the observed locations of events or items of interest.

Year 1 Year 2Figure 1: Campylobacter cases in Manawatu.

Figure 1 shows the pattern of human campylobacter infections inthe Manawatu region of New Zealand recorded in two di↵erentyears [55, 38]: spatial maps of disease cases are important forsurveillance of public health risk [41, 31, 44, 23, 73, 38]. In otherexamples, the disposition of artefacts in an ancient cemetery couldreflect social strata [67], and the spatial pattern of alarm calls ofcapuchins in a jungle may reflect social organisation [22].

Statistical methodology for spatial point patterns can extract this information e�ciently andrigorously [9, 30, 40, 53]. The observed pattern is treated as the outcome of a random spatial pointprocess [20]. Statistical methodology for point processes has become a highly active research topic,leveraging recent advances in statistical theory, algorithms and computing [9, 53, 54].

This project addresses unsolved, fundamental challenges in this methodology, with importantpractical consequences. The foremost problem is unidentifiability. An apparent cluster of diseasecases, for example, could be attributable either to a common systemic cause (such as a commonsource of infection) or to stochastic dependence between the points (such as contagion betweeninfected cases) [44, 43, 29, 55]. Distinguishing between these competing explanations is vital forunderstanding the disease aetiology and public health risk [31, 39]. Yet, for half a century, thishas been thought to be fundamentally impossible. The standard statistical approach is to formu-late a maximal model that includes all potential sources of variability, then use formal statisticalprocedures to decide which components of the model are needed to explain the data. But, by afamous result of Bartlett [16], point process models that include both deterministic heterogene-ity and stochastic dependence may be technically “unidentifiable” or “confounded”, in that themodel parameters cannot be determined from observation. The two competing explanations yieldidentical predictions, and therefore cannot be distinguished, even with infinite amounts of data.

In elementary statistics, unidentifiability is a defect of the model used, and can often be avoidedsimply by altering the model or the experiment. For spatial point processes, however, it hasproved extremely di�cult to construct models that simultaneously incorporate both deterministicheterogeneity and stochastic dependence, are identifiable and tractable for statistical inference,and are realistic descriptions of the data. Progress has been achieved only in specific applicationfields, for example in seismology, after decades of research [56, 57].

Faced with this apparent impasse, research on statistical methodology for spatial point patternshas essentially evolved as two separate streams, one focussed on deterministic trend, and the otheron dependence between points. Recent progress in smoothing techniques, such as kernel andspline methods (a key research area of PI Davies and AI Hazelton), now permits estimation of thedeterministic trend in a highly flexible and computationally e�cient manner [28, 34, 70, 71, 18,21, 23, 22, 26], but this work typically assumes independence between points. Progress in spatialpoint process modelling and statistical methodology now permits a rich variety of representationsof inter-point dependence (a specialty of AI Baddeley) [19, 6, 52, 53, 14, 9], but typically assumesthe deterministic trend is known or is rigidly parametrised. The fundamental weakness remainsthat deterministic trend and stochastic inter-point dependence cannot both be subjected to a highlevel of scrutiny in the same analysis. Statistical tools for spatial point patterns are thereforecurrently unsatisfactory for many important and relevant research questions.

Recent findings o↵er a glimmer of hope. In time series analysis (where unidentifiability is also along-standing problem) researchers have been able to disentangle global trend from autocorrelationby constraining the model [37, 59, 45, 36]. For spatial point patterns, the task is more complex,but a similar strategy has been successful in some very specific applications [31, 46, 51, 24]. Theseresults are limited, but they reveal the potential scope, power and utility of this general approach.

2B. OVERALL AIM OF THE RESEARCHWe aim to develop a new generation of statistical methods for the analysis of spatial point patterns,that can simultaneously support detailed inference about deterministic and stochastic sources of

ProposalStandard


InitialsTM


PanelMIS


variation. These methods will combine the best modern techniques for modelling deterministictrend and stochastic dependence, in a ‘hybridised’ statistical modelling framework. The resultingmethods will enable a much more searching analysis of important scientific questions about spatialpoint pattern data, for example in spatial epidemiology.

Objective 1: Development of identifiable spatial point process models. A model isunidentifiable when its statistical properties at all spatial scales can be explained equally well bydeterministic or stochastic e↵ects [16]. This suggests that identifiability could be recovered byconstraining the model so that the deterministic e↵ects operate at (say) a larger scale than thestochastic e↵ects. This strategy has been used successfully (albeit in very specific fashions) in timeseries analysis [37, 59, 45, 36] and in spatial point pattern analysis [31, 46, 51, 24].

Our objective is to develop flexible model classes for spatial point processes in which deter-ministic and stochastic e↵ects are combined and yet are identifiable. This requires theoretical,computational and practical developments. Tools for constructing ‘hybrid’ point process models[15, 31] will ensure that the models are well-defined and tractable. Model assumptions about thespatial scale of the deterministic and stochastic e↵ects may either be stated explicitly (such asconstraints on the derivative of the point process intensity) or incorporated implicitly (throughthe choice of model-fitting criteria such as spatial penalized likelihood [63, 71, 50, 32] and localcomposite likelihood [3, 12, 35]) or encoded as prior information in a Bayesian analysis [69, 68].

Objective 2: Implement and assess practical methods of inference. We will developa novel library of tools necessary to permit researchers to analyse real-world data. There will bea strong focus on methods for model fitting. We will investigate the theoretical properties of ourestimators within the new modelling frameworks; for example, examining how the properties ofkernel estimators of intensity are impacted by assumptions on ranges of e↵ect. We will work oncomputationally e�cient algorithms to ensure prompt calculation of estimates at a high level ofspatial resolution. Finally, we will investigate diagnostic methods to assess adequacy of model fit,building on the seminal work of the AIs [14] in spatial statistics.

Objective 3: Extend developments to point patterns on non-standard domains.There is an increasing literature dealing with points observed on special domains which requireunique treatment. Good examples are given by patterns observed on linear networks [17, 58] oron the surface of a sphere [33, 42], including recent work involving PI Davies and AI Baddeley[48, 61]. Little is known about how one should combine deterministic and stochastic e↵ects inthese instances, and how model fitting is a↵ected. In response, we will assess the unique modellingrequirements such data presents, and examine the feasibility of extending the work on the first twoobjectives to these situations.

Throughout the project, particularly Objectives 2 and 3, computational implementations willbe necessary to test and appraise the developed methodology. Thus, a goal parallel to our entirebody of work will be to produce corresponding software for general consumption by the researchcommunity, ensuring accessibility of our new, flexible statistical techniques. Ideal vehicles arealready available, by way of AI Baddeley’s prominent R [60] package spatstat [13, 9], as well asthe sparr package by PI Davies and AI Hazelton [25, 27].

2C. PROPOSED RESEARCHSuppose we observe a point pattern x = {x1, . . . , xn}, taken to be a realisation of the pointprocess X defined within some spatial domain W ⇢ R2. Our proposed research is centred onmodels capable of combining deterministic and stochastic e↵ects to describe the data at hand.

To model “clustering” or positive association one may use a Cox point process model [19, 53] inwhich, conditional on the realisation (x) of an external stochastic process (x), the point processX is a Poisson process [9, 30] with intensity function of the form

�X| (x) = ⌫(x) (x), x 2 W, (1)where ⌫ is a deterministic “trend” function that would be modelled using advanced smoothingtechniques. This modelling scheme has been used in several application-specific examples e.g.[31, 51]. For identifiability we require E[ (x)] ⌘ 1 so that the marginal intensity of X is �X(x) =⌫(x). The stochastic component plays the role of a random e↵ect [9, p. 450], and gives rise to

ProposalStandard


InitialsTM


PanelMIS


positive spatial correlation; it is governed by a vector of correlation parameters ✓. It is technicallychallenging to construct new models for which have desired properties such as a specified spatialcorrelation structure. Consequently the model for is usually selected from a menu of formulationswhich have been theoretically analysed: reasonable choices for include log-Gaussian randomfields, shot-noise fields and random mosaic models.

Figure 2 shows a synthetic example in W = [0, 1]2. The left panel was generated from a purely λX(x) = ν(x) λX | Ψ(x) = ν(x) ψ(x)

Figure 2: Artificially generated data used to illustrate adeterministic intensity (left) and a random intensity

combining deterministic and stochastic e↵ects (right).

deterministic intensity function �X(x) = ⌫(x); this iscontrasted with a dataset generated using the multiplica-tive construction of (1) using the same deterministic func-tion ⌫(x). There we see smaller-scale sporadic clusteringin tandem with the overall heterogeneity—in practice,this would be the nature of the data we observe (cf. Fig-ure 1). As the volume of data increases we can hope toobtain an ever more accurate estimate of the marginal in-tensity �X| (x), but that will generally not provide any

means of separating the relative contributions of ⌫ and .Complicated forms of dependence between points can also be described using Gibbs point pro-

cess models [62, 6, 47] which can best be formulated and fitted using their (Papangelou) conditionalintensity �X(u | x), u 2 W . By the Mobius inversion formula, this can be factorised as

�X(u | x) = �(u)G(u | x), u 2 W, (2)

where �(u) is the deterministic trend and G(u | x) contains only interaction terms. This canbe treated analogously to (1) by modelling � using advanced smoothing techniques, while Gcan only take certain forms which are known (from previous research) to satisfy the technicalrequirements. An important di↵erence is that the trend � and intensity �X are no longer equal,but are approximately related by a functional equation [7].

Historically it has been important that the terms which induce stochastic dependence, (x) in(1) and G(u | x) in (2), cannot be chosen willy-nilly because they must satisfy technical constraints.They have usually been selected from a short menu of models that have previously been constructedand checked for validity in the literature. In both cases, products of terms chosen from the menuare also valid, so that (x) could be replaced by 1(x) . . . m(x) where j(x), j = 1, . . . ,m areprocesses known to satisfy the technical conditions, and similarly G(u | x) may be validly replacedby a product of such terms [15]. This increases the scope of modelling considerably.

Application-specific examples in the literature reveal the extraordinary flexibility made possibleby hybrid deterministic-stochastic modelling frameworks such as (1); the intuition behind whichis supported by the real-world context of the data at hand [14, 31, 46, 51]. This in turn shines alight on the potential for significant advances in spatial point process statistics. The fundamentalchallenge, however, is to be able to separately capture and hence reliably model both ⌫(x) and theproperties of the stochastic driving mechanism (x) given an observed dataset. Work to date onthis di�cult problem has remained ad-hoc and severely limited in a more general sense.

Our proposed research thus seeks to develop a new generation of statistical methods thatcircumvent the inherent unidentifiability by placing informed, readily estimable constraints on thecomponent processes. This will overcome a decades-long stalemate on progress and allow us novel,highly flexible means to address complex research questions arising in a variety of disciplines.

Objective 1: Development of identifiable spatial point process models. Strategiesfor proceeding are best posed by first considering a particular model design, followed by specifica-tion of the nature of the component processes. For the sake of exposition, let us remain with theformulation laid out by (1). Diggle et al. [31] used a spatiotemporal version of this design for theanalysis of gastrointestinal infections in a region of the UK. They addressed the issue of identifia-bility between the deterministic and stochastic elements by e↵ectively making assumptions aboutthe smoothness of ⌫. In more detail, they estimated ⌫ by (spatially adaptive) kernel smoothingwith a subjectively chosen global bandwidth h, and then accounted for the residual variation inthe intensity using a log-Gaussian Cox process [52] as the stochastic driving mechanism . In this

ProposalStandard


InitialsTM


PanelMIS


sense, arises as a realisation of where

(x; ✓) = exp{Y (x; ✓)}; x 2 W ⇢ R2, (3)

where Y is taken to be a stationary, isotropic Gaussian random field with parameter ✓ = {µ, �2,�}in which we define a mean µ, variance �2, and correlation scale parameter �. With the kernelestimate of the deterministic trend treated as known, the parameters necessary for (3) were subse-quently estimated using minimum contrast methods [53, 30]. An approach of this sort is pragmaticand can be e↵ective, but it raises some theoretical di�culties. If the bandwidth follows the usualasymptotic regimen for consistency of the kernel estimator, then we may expect ⌫ ! �X| asn ! 1 [70], technically obviating the need for . On the other hand, if we place some lowerbound on the bandwidth h, then ⌫ will not be a consistent estimator in general.

As part of our proposed work we will pursue the idea that stochastic variation may be dis-tinguished from deterministic trend by consideration of scale. One approach is to operate underformal assumptions on the curvature of ⌫. We anticipate this will lead to theoretical identifiabilityfor some classes of stochastic models , if suitable constraints are placed on the range of spatialcorrelation. Equivalently we may look at contributions of ⌫ and in the frequency domain, ex-plaining the high frequency variations through the stochastic element. The idea is rather general,and could for instance be used to resolve the identifiability problem studied by Bartlett [16].

There is some precedent in the literature for the above strategy. In kernel density estimation,guidelines to optimal bandwidth selection can be linked to the estimated smoothness of the targetfunction via its second derivative [64, 66]. An adaptation of such an approach in the presence ofspatial dependence, by studying these properties of a kernel estimate of ⌫, is one avenue of pursuit.

Moving away from kernel smoothing, note that model identifiability will also typically be as-sured if the deterministic trend is specified parametrically. In principle this means we can representthat term using some kind of bivariate spline with a fixed number of knots [71]. We will addresstwo questions that naturally arise. First, what can be said about identifiability under an asymp-totic regimen in which the number of knots increases with the n, but at a (much) slower rate?Second, how much flexibility can we hope to achieve in practice from a spline estimate of trendwhile permitting reasonable estimates of model parameters?

A di↵erent approach is to circumvent problems of identifiability through use of local likelihoodmethods, recently pioneered by AI Baddeley [3]. This involves creation of a pastiche of local modelsbuilt up from a pre-specified “template” model which itself does not exhibit spatial inhomogeneity.Each localised model is fitted using likelihood functions weighted to the data in the neighbourhoodof the estimation point in question. Issues of scale therefore remain, and are bound up withcalibration of the instrumental kernel weights. We will seek to further develop this approach.

Objective 2: Implement and assess practical methods of inference. The work underthis objective follows naturally from theoretical developments under the first objective. In essence,we seek to ensure that theoretical knowledge is e↵ectively translated to practical methodologiesfor model fitting and diagnosis.

If identifiability is enabled through placing restrictions on the curvature of ⌫, then a naturalapproach is to estimate this component of the intensity through a kernel estimate with a corre-sponding kind of constraint. However, we face theoretical challenges because the curvature of thekernel estimate ⌫ will be an amalgam of the true curvature of ⌫ with additional high frequency‘wiggles’ corresponding to noise in the dataset at hand. It follows that applying a theoretical boundon the curvature of ⌫ to the estimator ⌫ will not necessarily guarantee consistent estimation. Someinsight into this problem exists for univariate smoothing [65]; extending these results to the bi-variate setting will permit design of a consistent kernel estimator of ⌫ under curvature constraints.We will then seek to develop data-driven bandwidth selectors for use in this constrained setting.Finally, we will also explore the extension of these results from fixed bandwidth kernel estimationto spatially adaptive estimation (e.g. based on the methods of [1]); shown to be well-suited topoint pattern data [23, 22]. In principle the utilisation of these constrained kernel estimators willallow us to estimate consistently the parameters ✓ of the residual stochastic process . We willconduct numerical experiments to assess the practicability of this approach for single instances of

ProposalStandard


InitialsTM


PanelMIS


point patterns (see relevant commentary on the di�culties faced in general e.g. [49]).Turning to spline methods, we note that while knot selection (including both the number and

position of knots) is often relatively unimportant for univariate splines, this choice is likely to befar more important in the 2D setting [63, 71]. The curse of dimensionality means that a dense gridof knots locations will not be practicable, and even with smaller numbers of knots, optimal choicebetween knot arrangements will be challenging. We will examine these issues in the presence ofvarious driver processes , taking account of estimation accuracy and computational expense.

Local likelihood methods have the powerful advantage that they are easily deployed in appli-cations: they do not require the construction and theoretical analysis of bespoke models. Existingtheory, methodology and computational techniques relevant to the template model require onlyminor modification. The main disadvantage is that the final result is not fully specified in the fa-miliar sense of a “fitted model” [3]. Appropriate bandwidth selection for the kernel weights requiresattention, and we will seek to further develop such methods in line with theoretical advancements.

Diagnostic methods are important for the assessment of any statistical model. In the settingof spatial point processes, the work of the AIs [14, 4, 10] was seminal. However, for the hybridmodels at hand we will need to develop new tools, including analogues of residuals, leverage andinfluence, that are capable of indicating the extent to which deterministic and stochastic elementsof the model are uniquely determined by the data at hand.

Objective 3: Extend developments to point patterns on non-standard domains.Spatial point patterns are most commonly assumed to arise in continuous 2D space. How-ever, emerging research is highlighting the substantial challenges that arise when the patternarises on another domain, such as the surface of a sphere [33, 42] or a linear network [17, 58].

●

● ●●

● ●

●● ●

● ● ●●

●

●

●

●

●

●

●

●

●

●

●●●

●●

●

●

●

●

●

● ● ● ● ●●

●●

●●●●

● ● ● ●

● ●

●● ● ●

● ●● ●

●● ●

● ● ●●

●

●

●●

●●

● ●●

●

●●

●●●●●

●●● ● ● ● ●

●●●

●● ● ● ●

●●● ●

●

●

● ●●

●

●

●

●●●

●

●

●

Figure 3: Street crimes observed in an area of Chicago (left)and corresponding nonparametric intensity estimate (right;

line thickness proportional to intensity) using state-of-the-artmethodology.

The left panel of Figure 3 shows the spatial pattern ofstreet crimes reported in an urban area of Chicago, USA[2]. Here the crime locations are constrained to lie onthe street network. Similar examples include road tra�caccidents [61] and microscopic features on the dendritenetwork of a neuron [5].

The major challenge here is that the network is nothomogeneous, so there are no non-trivial homogeneousrandom processes on a network, and most of the standard recipes for model-building are notsuccessful [8]. A linear network is the simplest example of a “substrate” on which the data areconstrained to lie. This constraint a↵ects the entire approach to data analysis and completelyinvalidates many classical techniques which assume homogeneity. Even the simple task of non-parametric estimation of intensity is challenging on a linear network. Existing methods for 2Dpoint patterns cannot be transferred directly to a network; attempting to do so has led to falla-cious results [72] which were subsequently pointed out [48]. In recent work involving PI Daviesand AI Baddeley [48, 61] it is demonstrated that a 2D kernel intensity estimate of the points,renormalised by a convolution of the 2D kernel with arc-length measure on the network, leads toa statistically consistent estimator of intensity on the network. On the right of Figure 3 is such anestimate for the Chicago data.

Research on relevant statistical methods is still in its infancy for such problems, and no generalparametric modelling solutions presently exist, much less any hybridised deterministic-stochasticframeworks. However, the concept of combining the two sources of variation in modelling such apattern remains sound with similar scale-specific intuition. In the Chicago crimes example the fixedheterogeneity in the population in that urban area could be explained via a deterministic trend,with the stochastic component used to capture any more sporadic, localised crime tendencies.

In our final task of the proposed research we aim to make a substantial contribution to thisemerging area of study. We will use hybridisation and localisation strategies to construct pointprocess models on a linear network which are amenable to analysis while being su�ciently flexibleto model real data. The modelling and estimation tools developed in Objectives 1 and 2 will beenlisted and tested in this challenging context.

ProposalStandard


InitialsTM


PanelMIS


2E. REFERENCES[1] Abramson IS (1982) On bandwidth variation in kernel estimates—a square root law, Annals

of Statistics 10 1217-1223.[2] Ang QW,Baddeley A, Nair G (2012) Geometrically corrected second-order analysis of events

on a linear network, with applications to ecology and criminology, Scandinavian Journal of

Statistics 39 591-617.[3] Baddeley A (2017) Local composite likelihood for spatial point processes, Spatial Statistics

22 261-295.[4] Baddeley A, Chang Y-M, Song Y, Turner R (2012) Residual diagnostics for covariate e↵ects

in spatial point process models. Journal of Computational and Graphical Statistics 22 886-905.[5] Baddeley A, Jammalamadaka A, Nair G (2014) Multitype point process analysis of spines

on the dendrite network of a neuron, Applied Statistics 63 673-694.[6] Baddeley A, Møller J (1989) Nearest-neighbour Markov point processes and random sets,

International Statistical Review 2 89-121.[7] Baddeley A, Nair G (2012) Fast approximation of the intensity of Gibbs point processes.

Electronic Journal of Statistics 6, 1155-1169.[8] Baddeley A, Nair G, Rakshit S, McSwiggan G (2017) “Stationary” point processes are

uncommon on linear networks. STAT 6, 68-78.[9] Baddeley A, Rubak E, Turner R (2015) Spatial Point Patterns: Methodology and Applica-

tions With R, Chapman and Hall/CRC Press, UK.[10] Baddeley A, Rubak E, Turner R (2018) Leverage and influence diagnostics for Gibbs spatial

point processes. Spatial Statistics 29, 15-48.[11] Baddeley A, Møller J, Waagepetersen R (2000) Non- and semi-parametric estimation of

interaction in inhomogeneous point patterns, Statistica Neerlandica 54 329-350.[12] Baddeley A, Turner R (2000) Practical maximum pseudolikelihood for spatial point patterns

(with discussion), Australian and New Zealand Journal of Statistics 42 283-322.[13] Baddeley A, Turner R (2005) spatstat: an R package for analyzing spatial point patterns,

Journal of Statistical Software 12 1–42.[14] Baddeley A, Turner R, Møller J, Hazelton ML (2005) Residual analysis for spatial point

processes (with discussion), Journal of the Royal Statistical Society Series B 67 617-666.[15] Baddeley A, Turner R, Mateu J, Bevan A (2013) Hybrids of Gibbs point process models

and their implementation, Journal of Statistical Software 55, 11, 1–43.[16] Bartlett MS (1964) The spectral analysis of two-dimensional point processes, Biometrika 51

299–311.[17] Borruso G (2008) Network density estimation: A GIS approach for analysing point patterns

in a network space, Transactions in GIS 12 377-402.[18] Botev Z, Grotowski J, Kroese D (2010) Kernel density estimation via di↵usion, Annals of

Statistics 38 2916-2957.[19] Cox DR (1955) Some statistical models related with series of events, Journal of the Royal

Statistical Society Series B 17 129-164.[20] Daley DJ and Vere-Jones D (2003) An Introduction to the Theory of Point Processes, Volume

I: Elementary Theory and Methods (2nd. Ed.), Springer, USA.[21] Davies TM (2013) Jointly optimal bandwidth selection for the planar kernel-smoothed

density-ratio, Spatial and Spatio-temporal Epidemiology 5 51-65.[22] Davies TM, Baddeley A (2018) Fast computation of spatially adaptive kernel estimates,

Statistics and Computing 28 937-956.[23] Davies TM, Hazelton ML (2010) Adaptive kernel estimation of spatial relative risk, Statis-

tics in Medicine 29 2423-2437.[24] Davies TM, Hazelton ML (2013) Assessing minimum contrast parameter estimation for

spatial and spatiotemporal log-Gaussian Cox processes, Statistica Neerlandica 67 355-389.[25] Davies TM, Hazelton ML, Marshall JC (2011) sparr: Analyzing spatial relative risk using

fixed and adaptive kernel density estimation in R, Journal of Statistical Software 39 1-14.[26] Davies TM, Lawson AB (2019) An evaluation of likelihood-based bandwidth selectors for

ProposalStandard


InitialsTM


PanelMIS


spatial and spatiotemporal kernel estimates, Journal of Statistical Computation and Simula-

tion 89 1131-1152.[27] Davies TM, Marshall JC, Hazelton ML (2018) Tutorial on kernel estimation of continuous

spatial and spatiotemporal relative risk, Statistics in Medicine 37 1191-1221.[28] Diggle PJ (1985) A kernel method for smoothing point process data, Applied Statistics 34

138-147.[29] Diggle PJ (1990) A point process modelling approach to raised incidence of a rare phenomenon

in the vicinity of a prespecified point, Journal of the Royal Statistical Society Series A 153349-362.

[30] Diggle PJ (2014) Statistical Analysis of Spatial and Spatiotemporal Point Patterns, 3rd ed,Chapman and Hall/CRC Press, USA.

[31] Diggle PJ, Rowlingson B, Su TL (2005) Point process methodology for on-line spatio-temporaldisease surveillance, Environmetrics 16 423-434.

[32] Fahrmeir L, Kneib T (2011) Bayesian smoothing and regression for longitudinal, spatial and

event history data, Oxford University Press, UK.[33] Fisher N, Lewis T, Embleton B (1993) Statistical Analysis of Spherical Data, Cambridge

University Press, Cambridge, UK.[34] Green PJ, Silverman BW (1993) Nonparametric regression and generalized linear models: a

roughness penalty approach, CRC Press, USA.[35] Guan Y, Jalilian A, Waagepetersen R (2015) Quasi-likelihood for spatial point processes,

Journal of the Royal Statistical Society Series B 77 677–697.[36] Guerrero VM, Cortes Toto D, Reyes Cervantes HJ (2018) E↵ect of autocorrelation when

estimating the trend of a time series via penalized least squares with controlled smoothness,Statistical Methods and Applications 27 109-130.

[37] Hart JD (1991) Kernel regression estimation with time series errors, Journal of the Royal

Statistical Society Series B 53 173-187.[38] Hazelton ML (2017) Testing for changes in spatial relative risk, Statistics in Medicine 36

2735-2749.[39] Hocking B, Gordon IR, Grain HL, Hatfield GE (1996) Cancer incidence and mortality and

proximity to TV towers, Medical Journal of Australia 165 601-605.[40] Illian J, Penttinen A, Stoyan H, Stoyan D (2008) Statistical Analysis and Modelling of Spatial

Point Patterns, Wiley, UK.[41] Kelsall JE, Diggle PJ (1995) Non-parametric estimation of spatial variation in relative risk,

Statistics in Medicine 14, 2335-2342.[42] Lawrence T, Baddeley A, Milne RK, Nair G (2016), Point pattern analysis on a region of a

sphere, Stat 5 144-157.[43] Lawson AB (2018) Bayesian Disease Mapping: Hierarchical Modeling in Spatial Epidemiology

(3rd. Ed.), Chapman and Hall, USA.[44] Lawson AB, Banerjee S, Haining RP, Ugarte MD (eds.) (2018) Handbook of Spatial Epidemi-

ology, Chapman and Hall, USA.[45] Lee YK, Mammen E, Park BU (2010) Bandwidth selection for kernel regression with correlated

errors, Statistics 44 327-340.[46] Liang S, Carlin BP, Gelfand AE (2009), Analysis of Minnesota colon and rectum cancer point

patterns with spatial and nonspatial covariate information, Annals of Applied Statistics 3943-962.

[47] van Lieshout, MNM (2000) Markov Point Processes and their Applications, Imperial CollegePress, London.

[48] McSwiggan G, Baddeley A, Nair G (2016) Kernel density estimation on a linear network,Scandinavian Journal of Statistics 44 324-345.

[49] Micheas AC (2019) Cox point processes: why one realisation is not enough, InternationalStatistical Review [in press].

[50] Mohler GO, Bertozzi AL, Goldstein TA, Osher SJ (2011) Fast TV regularization for 2Dmaximum penalized likelihood estimation, Journal of Computational and Graphical Statistics

ProposalStandard


InitialsTM


PanelMIS


20 479-491.[51] Møller J, Dıaz-Avalos C (2010) Structured spatio-temporal shot-noise Cox point process mod-

els, with a view to modelling forest fires, Scandinavian Journal of Statistics 37 2-25.[52] Møller J, Syversveen AR, Waagepetersen R (1998) Log Gaussian Cox processes, Scandinavian

Journal of Statistics 25 451-482.[53] Møller J, Waagepetersen R (2004) Statistical Inference and Simulation for Spatial Point Pro-

cesses, Chapman and Hall, USA.[54] Møller J, Waagepetersen R (2017) Some recent developments in statistics for spatial point

patterns, Annual Review of Statistics and Its Applications, 4 317-342.[55] Mullner P, Shadbolt T, Collins-Emerson JM, Midwinter AC, Spencer SE, Marshall JC, Carter

P, Campbell D, Wilson D, Hathaway S, Pirie R, French NP (2010) Molecular and spatial epi-demiology of human campylobacteriosis: source association and genotype-related risk factors,Epidemiology and Infection 138 1372-1383.

[56] Ogata Y (1998) Space-time point process models for earthquake occurrences, Annals of the

Institute of Statistical Mathematics 50 379-402.[57] Ogata Y, Zhuang J (2006) Space-time ETAS models and an improved extension, Tectono-

physics 413 13-23.[58] Okabe A, Sugihara K (2012) Spatial Analysis Along Networks, John Wiley and Sons, USA.[59] Opsomer J, Wang Y, Yang Y (2001) Nonparametric regression with correlated errors, Statis-

tical Science 16 134-153.[60] R Core Team (2019) R: A language and environment for statistical computing, R Foundation

for Statistical Computing, Vienna, Austria. URL: https://www.R-project.org/.[61] Rakshit S, Davies TM, Moradi MM, McSwiggan G, Nair G, Mateu J, Baddeley A (2019)

Fast kernel smoothing of point patterns on a large network using 2D convolution, InternationalStatistical Review [in press].

[62] Ripley BD, Kelly FP (1977) Markov point processes Journal of the London Mathematical

Society, 15 188-192.[63] Ruppert D, Wand MP, Carroll RJ (2003) Semiparametric Regression, Cambridge University

Press, UK.[64] Silverman BW (1978) Choosing the window width when estimating a density, Biometrika 65

1-11.[65] Silverman BW (1978) Weak and strong uniform consistency of the kernel estimate of a density

and its derivatives, Annals of Statistics 6 177-184.[66] Silverman BW (1986) Density Estimation for Statistics and Data Analysis, Chapman and

Hall, USA.[67] Smith BA, Davies TM, Higham CFW (2015) Spatial and social variables in the Bronze Age

phase 4 cemetery of Ban Non Wat, Northeast Thailand, Journal of Archaeological Science:Reports 4 362-370.

[68] Taylor BM, Davies TM, Rowlingson BS, Diggle PJ (2015) Bayesian inference and dataaugmentation schemes for spatial, spatiotemporal and multivariate log-Gaussian Cox processesin R, Journal of Statistical Software 63 1-48.

[69] Taylor BM, Diggle PJ (2014) INLA or MCMC? A tutorial and comparative evaluation forspatial prediction in log-Gaussian Cox processes, Journal of Statistical Computation and Sim-

ulation 84 2266-2284.[70] Wand MP, Jones MC (1995) Kernel Smoothing, Chapman and Hall, USA.[71] Wood SN (2003) Thin plate regression splines, Journal of the Royal Statistical Society Series

B 65 95-114.[72] Xie Z, Yan J (2008) Kernel density estimation of tra�c accidents in a network space, Com-

puters, Environment and Urban Systems 32 396-406.[73] Zhang ZJ, Davies TM, Gao J, Wang Z, Jiang QW (2013) Identification of high-risk regions

for schistosomiasis in the Guichi region of China: an adaptive kernel density estimation-basedapproach, Parasitology 140 868-875.

ProposalStandard


InitialsTM


PanelMIS


2F. TIMETABLE

Number of Years (maximum of 3): 3Year 1: 2020

• Investigation of spatial identifiability for models of point process data.– Derive general properties of models under specific constraints on deterministic and

stochastic components.

– Conduct ‘proof-of-concept’ empirical tests/simulations for the di↵erent formulationsinvestigated above to assess ‘practical identifiability’ i.e. what we might hope to learnfrom individual datasets.

• Develop formal model designs for the most promising identifiable frameworks.• PI Davies and AI Hazelton visited by AI Baddeley at the University of Otago. Plan and draft2 manuscripts dealing with the di↵erent aspects of the theory and and practical consequencesof constraining individual model components. Recruit PhD student.

Year 2: 2021

• A concentration on practical methods of model fitting and inference for newly developedhybrid spatial point process models.

– Develop data-driven ways to secure constraints on the deterministic and stochasticcomponents as required.

– Develop estimation methods for deterministic components in the presence of spatialdependence and vice-versa.

– Empirically investigate estimation techniques.• Develop diagnostic tools for newly developed models, to assess their fit to data. Assessestimation methods, plots, and ease of interpretation.

• Further visit between the three named members of the research team (venue TBD). Plan anddraft 2-3 manuscripts on practical fitting methods, diagnostic investigation, and applications.Recruit 2-year postdoctoral fellow.

Year 3: 2022

• Development of hybridised modelling techniques for point patterns on linear networks.– Define stochastic processes on networks; assess properties.

– Building on recent work for smoothing network data, design possible modelling frame-works for linear network data combining deterministic and stochastic e↵ects.

– Test models using empirical and real-world data.• Consider potential for similar models for alternative domains, such as point patterns observedon the surface of a sphere.

• Further visit between the three named members of the research team (venue TBD). Plan1-2 manuscripts with particular focus on modelling point pattern data observed on linearnetworks.

In tandem with the above, we will implement methodological advancements in the open-sourcesoftware package R, extending the functionality of existing R libraries developed by the researchteam. We recognise, and are particularly motivated by, the need for accessibility of cutting-edgestatistical methods for inference to the wider research community. We expect this to serve as anexcellent avenue for the drafting of additional ‘tutorial-style’ manuscripts and software vignettesfor applied researchers and other users, thereby raising the profile of New Zealand’s expertise inthese areas. Furthermore, progress will be reported on regularly at both local and internationalstatistics conferences.

2G. ROLES AND RESOURCESThe proposed programme of research seeks to bridge the critical gap that currently exists betweenflexible smoothing methods for estimation of fixed components of trend on the one hand, and

ProposalStandard


InitialsTM


PanelMIS


stochastic models to describe point-to-point interaction on the other. This is perfectly reflectedin the composition of the research team, which brings together two of Australasia’s top statisticalresearchers in spatial point processes (AI Baddeley) and kernel smoothing (AI Hazelton), alongwith a 2-year postdoctoral fellow and a 3-year Ph.D. student, under the leadership of emergingresearch heavyweight PI Davies. This trans-Tasman group has the experience and dynamism toconduct world leading research, firmly positioning New Zealand at the frontier of spatial statistics.

PI Davies has established himself as one of New Zealand’s leading young statisticians,with over two dozen papers in high-impact statistics and applications journals (e.g. Stat. Med.,Ann. Appl. Stat., Stat. Comp.) and two early career research awards. He has expertise in bothspatial statistics and smoothing methods, including methodological development, computationalimplementation and application. With 0.25 FTE, he will lead the project, contribute to allareas of research, and be the primary supervisor of the postdoctoral and doctoral researchers. Hisunifying role will serve to strengthen the intersection of knowledge shared by the two experiencedAIs.

AI Baddeley is a Distinguished Professor and recognised international researcher in the fieldof spatial point process statistics, with scores of publications in top statistics journals (e.g. Ann.Stat., JRSS B) and numerous scientific prizes and a�liations. Building on this vast expertise,AI Baddeley will contribute in particular to theoretical developments related to estimation ofinteraction e↵ects and associated software development, with a proposed 0.1 FTE. Of note is hislead authorship of the spatstat software, which provides a superb dissemination vehicle for noveldevelopments of the project: The accompanying 2005 paper by Baddeley and Turner (ref. [13] inSection 2E) has now been cited well over 1,000 times.

AI Hazelton is a prolific international researcher in kernel density estimation and relatedsmoothing problems. He has considerable knowledge of spatial statistics and associated applica-tions in spatial and spatio-temporal epidemiology. With a string of publications in leading statisticsjournals (e.g. JRSS B, Ann. Appl. Stat.), he has received the premier research prize of the NewZealand Statistical Association. Contributing 0.05 FTE, AI Hazelton will focus in particular onmethodological developments related to flexible nonparameteric trend estimation in the presenceof complex models for inter-point interaction.

Collaboration between the PI and AIs will be further strengthened through the ‘glue’ pro-vided by the postdoctoral fellow and doctoral student, each at 1.0 FTE. Their specific roleswill depend in part on their own strengths and interests. Suitable candidates would have strongtheoretical and computational abilities, and an interest in spatial applications. While based pre-dominately with PI Davies at the University of Otago, both will also take opportunities to spendsignificant amounts of time in Western Australia working with Distinguished Professor Baddeleyand his research team. Their active support of the named investigators will lend particular strengthto the growth of New Zealand’s capacity for research into cutting-edge point process methodology.

The PI and AIs have a record of highly successful collaboration. Each pair has alreadypublished together in leading journals on various aspects of spatial statistics—Davies and Badde-ley (Stat. Comp., Int. Stat. Rev.); Davies and Hazelton (Stat. Med., Comput. Stat. Data Anal.);Baddeley and Hazelton (JRSS B, discussion paper)—though never as a group of three. Comingtogether on this project therefore represents both a natural progression to their individual workand an opportunity to forge an outstanding new tripartite research collaboration.

The physical resources of the proposed research project are minimal, aside from the requestedFTEs. The only other major requested resource will be funding for travel. At their respectiveinstitutions the investigators possess su�cient computational resources for the planned pursuits.Digital video conversations over the Internet will be scheduled regularly.

2H. ETHICAL OR REGULATORY OBLIGATIONSN/A

ProposalStandard


InitialsTM


PanelMIS


5. CURRICULUM VITAE AND PUBLICATIONS

PART 11a. Personal detailsFull name Title First name Second name(s) Family name

Dr Tilman Marcus DaviesPresent position Senior Lecturer in StatisticsOrganisation/Employer University of OtagoContact address Department of Mathematics & Statistics

PO Box 56Dunedin, New Zealand Post code 9054

Work telephone 03-479 7772 Mobile 021-165 2690Email [email protected] website http://www.stats.otago.ac.nz/?people=tilman davies

1b. Academic qualifications

2012: PhD Statistics; Massey University.2007: Bachelor of Science Honours (BScHons) in Statistics (First class); Massey University.2006: Bachelor of Computer and Mathematical Sciences (BCM) in Applied Statistics/German;University of Western Australia.

1c. Professional positions held

2018-present: Adjunct Research Fellow, Dept. of Mathematics and Statistics, Curtin Uni-versity, Australia.2017-present: Senior Lecturer in Statistics, University of Otago.2012-2016: Lecturer in Statistics, University of Otago.2009-2011: Graduate Assistant and Doctoral Student, Massey University.2008: Statistician, The EMMES Corporation (Rockville MD, USA).

1d. Present research/professional speciality

• Analysis of planar point patterns and spatial statistics• Kernel smoothing and density-ratios• Computational statistics, R programming• Biostatistical applications in geographical epidemiology and physiology

1e. Total years research experience: 10 years (incl. PhD)

1f. Professional distinctions and memberships (including honours, prizes, schol-arships, boards or governance roles, etc)

Grants/Funding2015: Awarded Marsden Fast-start Grant 15-UOO-092: ‘Smoothing and inference for pointprocess data with applications to epidemiology’ as PI.2014: University of Otago Research Grant (UORG): “Spatial Methods for Intensity Estima-tion and their Performance in Epidemiology” as PI.Awards/Scholarships2017: University of Otago Early Career Award for Distinction in Research.2014: Worsley Early Career Research Award (NZ Statistical Association).2009: Top Achiever’s Doctoral Scholarship (Bright Future Scheme; TEC, NZ).Postgraduate/Honours Supervision2019: Morshadur Rahman, PhD (primary supervisor; expected start date: October).

ProposalStandard


InitialsTM


PanelMIS


2019-present: Anna Redmond, Honours (primary supervisor).2018-present: Megan Drysdale, Masters (co-supervisor).2018-present: Marilette Lotter, Honours (primary supervisor).2017-2018: Qing Ruan, Postgraduate Diploma (primary supervisor).2016-2018: Baylee Smith, Masters (co-supervisor).2015: Patrick Brown, Postgraduate Diploma (primary supervisor).2014: Baylee Smith, Honours (primary supervisor).2013: Claire Flynn, Honours (primary supervisor).Conference/Workshop Attendance2018: Australian Statistical Conference, Melbourne (ISCB/ASC 2018). Contributed talk.2017: Conference Board of Mathematical Statistics 2017 Regional Workshop: BayesianModeling for Spatial and Spatio-Temporal Data (CBMS 2017), Santa Cruz CA, USA.2016: Royal Statistical Society Conference (RSS 2016), Manchester, UK. Poster.2016: Institute of Mathematical Statistics Asia-Pacific Rim Meeting (IMS/APRM 2016), Chi-nese University of Hong Kong, Hong Kong. Invited speaker.2015: Otago International Health Research Network Conference (OIHRN 2015), Dunedin,NZ. Contributed talk.2015: Joint Statistical Meetings (JSM 2015), Seattle WA, USA. Contributed talk.2012-2016, 2018: Annual New Zealand Statistical Association Conference (NZSA/ORSNZ),main university centers around NZ. Contributed talks and poster presentations.2012: Australian and New Zealand Association of Clinical Anatomisits Conference (ANZACA2012), Sydney, Australia.Invited Collaboration and Seminars2018: Invited research visitor, Dept. of Biostatistics, UCLA, USA.2016-2018: Invited research visitor, Dept. of Mathematics and Statistics, Curtin University,Perth, Australia.2016, 2017: Invited research collaborator, Centre for Health Informatics Computing andStatistics (CHICAS), Lancaster University, UK.2011-2018: Local departmental statistics seminars, Dept. of Mathematics and Statistics,University of Otago, Dunedin, NZ.2011: Invited research visitor, Dept. of Mathematical Sciences, Aalborg University, Denmark.2011: Invited seminar, Dept. of Mathematics and Statistics, Lancaster University, UK.2011: Invited seminar, Dept. of Mathematics and Statistics, University of Jyvaskyla, Finland.Memberships2016-present: Accredited Statistician (AStat)—Statistical Society of Australia Inc. (SSAI).2012-present: Overseas member—Statistical Society of Australia Inc. (SSAI).2009-present: Regular member—New Zealand Statistical Association (NZSA).Professional/Outreach Roles2019: Chair, New Zealand Statistical Association Conference (NZSA2019).2018: Co-presenter with A. Baddeley and R. Turner for the ARC Centre for Excellence inMathematics and Statistics 3-day workshop on spatial statistics, University of Melbourne.2014-present: Director of Studies (100-level statistics: Dept. of Mathematics & Statistics,Otago).2013-present: Sole author/presenter of the annual 3-day University of Otago “Introductionto R” programming workshop (SWoPS 1: Statistics Workshops for Postgraduates and Staff).2013-present: Schools Liaison Officer, Dept. of Mathematics & Statistics, Otago.

1g. Total number of peer

reviewed publicationsand patents

Journalarticles

Books Book chap-ters, booksedited

Conferenceproceedings

Patents

25 1 0 1 0

ProposalStandard


InitialsTM


PanelMIS


PART 22a. Research publications and disseminationPeer reviewed journal articles

1. Davies TM, Lawson AB (2019) An evaluation of likelihood-based bandwidth selec-tors for spatial and spatiotemporal kernel estimates, Journal of Statistical Compu-

tation and Simulation 89(7) 1131-1152.

2. Davies TM, Schofield MR, Cornwall J, Sheard PW (2019) Modelling multilevel spa-tial behaviour in binary-mark muscle fibre configurations, Annals of Applied Statis-

tics [to appear] (doi: To be assigned).

3. Rakshit S, Davies TM, Moradi MM, McSwiggan G, Nair G, Mateu J, BaddeleyA (2019) Fast kernel smoothing of point patterns on a large network using 2Dconvolution, International Statistical Review [to appear] (doi: 10.1111/insr.12327).

4. Davies TM, Baddeley A (2018) Fast computation of spatially adaptive kernel esti-mates, Statistics and Computing 28(4) 937-956.

5. Davies TM, Flynn CR, Hazelton ML (2018) On the utility of asymptotic bandwidthselectors for spatially adaptive kernel density estimation, Statistics & Probability

Letters 138 75-81.

6. Davies TM, Marshall JC, Hazelton ML (2018) Tutorial on kernel estimation of con-tinuous spatial and spatiotemporal relative risk, Statistics in Medicine 37(7) 1191-1221.

7. Davies TM, Jones K, Hazelton ML (2016) Symmetric adaptive smoothing regimensfor estimation of the spatial relative risk function, Computational Statistics & Data

Analysis 101 12-28.

8. Davies TM, Sheard PW, Cornwall J (2016) Letter to the Editor: Comment onMakino et al. and observations on spatial modeling, Anatomical Science Inter-

national 91(4) 423-424.

9. Farrell S, Davies TM, Cornwall J (2015) Use of clinical anatomy resources bymusculoskeletal outpatient physiotherapists in Australian public hospitals: A cross-sectional study, Physiotherapy Canada 67(3) 273-279.

10. Fletcher JGR, Stringer MD, Briggs CA, Davies TM, Woodley SJ (2015) CT mor-phometry of adult thoracic intervertebral discs, European Spine Journal 24(10)2321-2329.

11. Smith BA, Davies TM, Higham CFW (2015) Spatial and social variables in theBronze Age phase 4 cemetery of Ban Non Wat, Northeast Thailand, Journal of

Archaeological Science: Reports 4(34) 362-370.

12. Taylor BM, Davies TM, Rowlingson BS, Diggle PJ (2015) Bayesian inferenceand data augmentation schemes for spatial, spatiotemporal and multivariate log-Gaussian Cox processes in R, Journal of Statistical Software 63(7) 1-48.

13. Cornwall J, Davies TM, Lees D (2013) Student injuries in the dissecting room,Anatomical Sciences Education 6(6) 404-409.

14. Davies TM (2013) Jointly optimal bandwidth selection for the planar kernel-smoothed density-ratio, Spatial and Spatio-temporal Epidemiology 5(1) 51-65.

ProposalStandard


InitialsTM


PanelMIS


15. Davies TM (2013) Scaling oversmoothing factors for kernel estimation of spatialrelative risk, Epidemiological Methods 2(1) 67-83.

16. Davies TM, Bryant DJ (2013) On circulant embedding for Gaussian random fieldsin R, Journal of Statistical Software 55(9) 1-21.

17. Davies TM, Cornwall J, Sheard PW (2013) Modelling dichotomously marked mus-cle fibre configurations, Statistics in Medicine 32(24) 4240-4258.

18. Davies TM, Hazelton ML (2013) Assessing minimum contrast parameter estima-tion for spatial and spatiotemporal log-Gaussian Cox processes, Statistica Neer-

landica 67(4) 355-389.

19. Taylor BM, Davies TM, Rowlingson BS, Diggle PJ (2013) lgcp - An R package forinference with spatial and spatiotemporal log-Gaussian Cox processes, Journal of

Statistical Software 52(4) 1-40.

20. Zhang ZJ, Davies TM, Gao J, Wang Z, Jiang QW (2013) Identification of high-risk regions for schistosomiasis in the Guichi region of China: an adaptive kerneldensity estimation-based approach, Parasitology 140(7) 868-875.

21. Zhang ZJ, Chen DM, Chen Y, Davies TM, Vaillancourt JP, Liu WB (2012) Risk sig-nals of an influenza pandemic caused by highly pathogenic avian influenza subtypeH5N1: Spatio-temporal perspectives, Veterinary Journal 192(3) 417-421.

22. Davies TM, Hazelton ML, Marshall JC (2011) sparr: Analyzing spatial relativerisk using fixed and adaptive kernel density estimation in R, Journal of Statistical

Software 39(1) 1-14.

23. Sanson RL, Harvey N, Garner MG, Stevenson MA, Davies TM, Hazelton ML,O’Connor J, Dube C, Forde-Folle KN, Owen K (2011) Foot-and-mouth diseasemodel verification and ‘relative validation’ through a formal model comparison, OIE

Scientific and Technical Review 30(2) 527-540.

24. Davies TM, Hazelton ML (2010) Adaptive kernel estimation of spatial relative risk,Statistics in Medicine 29(23) 2423-2437.

25. Hazelton ML, Davies TM (2009) Inference based on kernel estimates of the relativerisk function in geographical epidemiology, Biometrical Journal 51(1) 98-109.

Peer reviewed books

• Davies TM (2016) The Book of R: A First Course in Programming and Statistics,No Starch Press, San Francisco, USA; 832pp.

Refereed conference proceedings

• Davies TM, Sheard PW, Cornwall J (2013) Development of a novel statisticalmethod to test spatial distributions of skeletal muscle fiber types, In proceedings

of the 2012 Meeting of the Australian and New Zealand Association of Clinical

Anatomists (ANZACA); Clinical Anatomy 26, 641-660.

ProposalStandard


InitialsTM


PanelMIS




Prof. Martin Luke HazeltonPresent position Professor of StatisticsOrganisation/Employer Massey UniversityContact address Tennent Drive

Private Bag 11222Palmerston North Post code 4442

Work telephone 06-356 9099 Mobile 021-863 438Email [email protected] website(if applicable)

http://www.massey.ac.nz/emhazelto/

1b. Academic qualifications1993, D.Phil in Statistics, University of Oxford1989, BA Hons in Mathematics (First Class), University of Oxford

1c. Professional positions heldFrom September 2019, Professor of Statistics, University of Otago2017-Sept. 2019, Head of the Institute/School of Fundamental Sciences, Massey University2006-Sept. 2019, Chair of Statistics, Massey University1997-2006, Lecturer–Associate Professor in Statistics, University of Western Australia1994-1997, Lecturer in Statistical Science, University College London1993-1994, Research Officer in Transport Studies, University of Oxford1992-1993, Stipendiary Lecturer in Mathematics, Jesus College, University of Oxford

1d. Present research/professional specialityNetwork tomographyKernel smoothingSpatial statisticsStatistical modelling and inference for transport networksBiostatistics and statistical methods in epidemiology

1e. Total years research experience: 28 years


2017, Awarded Marsden grant 17-MAU-037, ‘Lattice polytope samplers: theory, methodsand applications’, as sole PI2016, Invited speaker at 4th IMS-APRM conference, The Chinese University of Hong Kong2015, AI on Marsden grant 15-UOO-092, ‘Smoothing and inference for point process datawith applications to epidemiology’2014, Awarded Littlejohn Research Award, the premier research award of the New ZealandStatistical Association2014, Awarded Marsden grant 14-MAU-017, ‘Modelling, inference and prediction for dy-namic traffic networks’, as sole PI

ProposalStandard


InitialsTM


PanelMIS


2014-2016, president of the New Zealand Statistical Association2014, Invited International Lecturer, Croucher Foundation Advanced Study Institute, HongKong University of Science and Technology2013, Keynote speaker at 2013 New Zealand Statistical Association/ Operations ResearchSociety of New Zealand conference, Hamilton2013, Invited speaker at HKUST Workshop on Day-to-Day Dynamical System Approach forModeling Transportation Systems, Hong Kong2013-2016, Associate Investigator on Australian Research Council Discovery Grant ‘Statis-tical methodology for events on a network, with application to road safety’2013-present, Associate Editor, Transportmetrica B: Transport Dynamics2012-present, Principal Investigator in the Infectious Disease Research Centre (www.idrec.ac.nz)2011-present, Theory and Methods Editor, Australian and New Zealand Journal of Statistics2011, plenary speaker at Statistical Concepts and Methods for the Modern World Confer-ence, Colombo, Sri Lanka2011, External academic on Graduating Year Review panel for Bachelor of MathematicalSciences at Auckland University of Technology2010-2014, Awards Committee Convenor, New Zealand Statistical Association2010-2011, Associate Editor, Australian and New Zealand Journal of Statistics2009, invited speaker at DADDY: Workshop on Day-to-day Dynamics for Transportation Net-works, Salerno, Italy2008-2016, Associate Editor, Journal of the Korean Statistical Association2008-2011, Awarded Marsden grant MAU0807, ‘New tools for statistical inference for network-based transportation models’, as sole PI2007-present, member of the Editorial Advisory Board, Transportation Research Part B2007, Grant assessor for Hong Kong City University2006, Invited speaker at joint Australian/New Zealand Statistics Conference, Auckland2006-present, Member of the New Zealand Statistical Association2006, Grant assessor for the Israel Science Foundation2005-2008, Grant assessor for Australian National Health & Medical Research Council2005, Author of read paper at Royal Statistical Society Research Meeting, See [28] in publi-cation list.2003-present, paper on plug-in bandwidth matrices for density estimation is most highly citedever in Journal of Nonparametric Statistics (Web of Science). See [39] in publication list2002, Invited speaker at 16th Australian Statistics Conference, Canberra, Australia2002-2004, President, West Australian Branch of the Statistical Society of Australia1998, Author of read paper at Royal Statistical Society General Meeting. See [48] in publi-cation list.1997-present, Supervised 13 PhD students, 9 as primary supervisor. Of the 10 to havecompleted to date, 7 secured academic posts1987-1989, Exhibitioner in mathematics, St Anne’s College, University of Oxford

1g. Total number of peerreviewed publicationsand patents

Journalarticles



Patents

82 0 9 5 0

ProposalStandard


InitialsTM


PanelMIS


PART 22a. Research publications and disseminationPeer reviewed journal articles (selected)

1. Liao, S. J., Marshall, J., Hazelton, M. L., & French, N. P. (2019). Extending statisti-cal models for source attribution of zoonotic diseases: a study of campylobacterio-sis. Journal of the Royal Society Interface, 16(150), 20180534.

2. Davies, T.M., Marshall, J.C. and Hazelton, M.L. (2018). Tutorial on kernel esti-mation of continuous spatial and spatiotemporal relative risk with accompanyinginstruction in R. Statistics in Medicine, 37, 1191-1221.

3. Davies, T.M., Flynn, C. and Hazelton, M.L. (2018). On the utility of asymptoticbandwidth selectors for spatially adaptive kernel density estimation. Statistics andProbability Letters, 138, 75-81.

4. Watling, D.P. and Hazelton, M.L. (2018). Asymptotic approximations of transientbehaviour for day-to-day traffic models. Transportation Research Part B, 118, 90-105.

5. Hazelton, M.L. (2017). Testing for changes in spatial relative risk. Statistics inMedicine, 36, 2735-2749.

6. Hazelton, M.L. and Bilton, T.P. (2017). Polytope samplers for network tomography.Australian and New Zealand Journal of Statistics, 59(4), 495-511.

7. Hazelton, M.L. and Cox, M.P. (2016). Bandwidth selection for kernel log-densityestimation. Computational Statistics and Data Analysis, 103, 56-67.

8. Hazelton, M.L. and Parry, K. (2016). Statistical methods for comparison of day-to-day traffic models. Transportation Research Part B, 92(A), 22-34.

9. Davies, T.M., Jones, K. and Hazelton, M.L. (2016). Symmetric adaptive smoothingregimens for estimation of the spatial relative risk function. Computational Statisticsand Data Analysis, 101, 12-18.

10. Pirikahu, S., Jones. G., Hazelton, M.L. and Heuer, C. (2016). Bayesian methodsof confidence interval construction for the population attributable risk from cross-sectional studies. Statistics in Medicine, 35 3117-3130.

11. Hazelton, M.L. (2015). Network tomography for integer-valued traffic. Annals ofApplied Statistics, 9(1), 474-506.

12. Fernando, W.T.P.S, Ganesalingam, S. and Hazelton, M.L. (2014). A comparisonof estimators of the geographical relative risk function. Journal of Statistical Com-putation and Simulation, 84(7), 1471-1485.

13. Davies, T.M and Hazelton, M.L. (2013). Assessing minimum contrast parameterestimation for spatial and spatiotemporal log-Gaussian Cox processes. StatisticaNeerlandica, 67(4), 355-389.

14. Parry, K. and Hazelton, M.L. (2013). Bayesian inference for day-to-day dynamictraffic models. Transportation Research Part B, 50, 104-115.

15. Parry, K. and Hazelton, M.L. (2012). Estimation of origin-destination matrices fromlink counts and sporadic routing data. Transportation Research Part B, 46, 175-188.

16. Hazelton, M.L. and Turlach, B.A. (2011). Semiparametric regression with shapeconstrained penalized splines. Computational Statistics and Data Analysis, 55,2871-2879.

ProposalStandard


InitialsTM


PanelMIS


17. Davies, T.M., Hazelton, M.L. and Marshall, J.C. (2011). sparr: Analyzing spatialrelative risk using fixed and adaptive kernel density estimation in R. Journal ofStatistical Software, 39, 1-14.

18. Hazelton, M.L. (2011). Assessing log-concavity of multivariate densities. Statisticsand Probability Letters, 81, 121-125.

19. Davies, T.M. and Hazelton, M.L. (2010). Adaptive kernel estimation of spatialrelative risk. Statistics in Medicine, 29, 2423-2437.

20. Hazelton, M.L. (2010). Bayesian inference for network-based modes with a linearinverse structure. Transportation Research Part B, 44, 674-685.

21. Hazelton, M.L. (2010). Statistical inference for transit system origin-destinationmatrices. Technometrics, 52, 221-230.

22. Marshall, J.C. and Hazelton, M.L. (2010). Boundary kernels for adaptive densityestimators on regions with irregular boundaries. Journal of Multivariate Analysis101, 949-963.

23. Hazelton, M.L. and Turlach, B.A. (2010). Semiparametric density deconvolution.Scandinavian Journal of Statistics 37, 91-108.

24. Hazelton, M.L. and Turlach, B.A. (2009). Nonparametric density deconvolution byweighted kernel estimators. Statistics and Computing, 19, 217-228.

25. Hazelton, M.L. and Marshall, J.C. (2009). Linear boundary kernels for bivariatedensity estimation. Statistics and Probability Letters, 79, 999-1003.

26. Hazelton, M.L. and Davies, T.M. (2009). Inference based on kernel estimates ofthe relative risk function in geographical epidemiology. Biometrical Journal, 51,98-109.

27. Hazelton, M.L. (2008). Statistical inference for time varying origin-destination ma-trices. Transportation Research Part B, 42, 442–452.

28. Hazelton, M.L. (2007). Bias reduction in kernel binary regression. ComputationalStatistics and Data Analysis, 51, 4393-4402.

29. Hazelton, M.L. and Turlach, B.A. (2007). Reweighted kernel density estimation.Computational Statistics and Data Analysis, 51, 3057-3069.

30. Baddeley, A., Turner, R., Moller, J. and Hazelton, M. (2005). Residual analysis forspatial point processes (with discussion). Journal of the Royal Statistical SocietySeries B, 67, 617-666. Read before the Royal Statistical Society on Wednesday22nd June 2005.

31. Duong, T and Hazelton, M.L. (2005). Cross-validation bandwidth matrices formultivariate kernel density estimation. Scandinavian Journal of Statistics, 32, 485-506.

32. Duong, T. and Hazelton, M.L. (2005). Convergence rates for unconstrained band-width matrix selectors in multivariate kernel density estimation. Journal of Multi-variate Analysis, 93, 417-433.

33. Gurrin, L.C., Scurrah, K. and Hazelton, M.L. (2005). Tutorial in biostatistics: Splinesmoothing with linear mixed models. Statistics in Medicine, 24, 3361-3381.

34. Hazelton, M.L. and Watling, D.P. (2004). Computation of equilibrium distributionsof Markov traffic assignment models. Transportation Science, 38, 331-342.

ProposalStandard


InitialsTM


PanelMIS


35. Hazelton, M.L. (2004). Density estimation from aggregate data. ComputationalStatistics, 19, 407-423.

36. Sircombe, K.N. and Hazelton, M.L. (2004). Comparison of detrital zircon agedistributions by kernel functional estimation. Sedimentary Geology, 171, 91-111

37. Hazelton, M.L. (2004). Density estimation from aggregate data. ComputationalStatistics, 19, 407-423.

38. Hazelton, M.L. (2003). A graphical tool for assessing normality. The AmericanStatistician, 57, 285-288.

39. Hazelton, M.L. (2003). Variable kernel density estimation. Australian and NewZealand Journal of Statistics, 45, 271-284.

40. Duong, T and Hazelton, M.L. (2003). Plug-in bandwidth selectors for bivariatekernel density estimation. Journal of Nonparametric Statistics, 15, 17-30.

41. Hazelton, M.L. (2003). Some comments on origin-destination matrix estimation.Transportation Research Part A, 37, 811-822.

42. Hazelton, M.L. (2003). Total travel cost in stochastic assignment models. Net-works and Spatial Economics, 3, 457-466.

43. Hazelton, M.L. (2002). Day-to-day variation in Markovian traffic assignment mod-els. Transportation Research Part B, 36, 637-648.

44. Hazelton, M.L. (2001). Estimation of origin-destination trip rates in Leicester. Jour-nal of the Royal Statistical Society, Series C (Applied Statistics), 50, 423-433.

45. Hazelton, M.L. (2001). Inference for origin-destination matrices: estimation, re-construction and prediction. Transportation Research Part B, 35, 667-676.

46. Hazelton, M.L. (2000). Marginal density estimation from incomplete bivariate data.Statistics and Probability Letters, 47, 75-84.

47. Hazelton, M.L. (2000). Estimation of origin-destination matrices from link flows onuncongested networks. Transportation Research Part B, 34, 549-566.

48. Broughton, J., Hazelton, M.L. and Stone, M. (1999). Influence of light-level onthe incidence of road casualties and the associated effect of summertime clockchanges. Journal of the Royal Statistical Society, Series A, 162, 137-175. Readbefore the Royal Statistical Society on 14 October 1998.

49. Hazelton, M.L. (1998). Bias annihilating bandwidths for kernel density estimationat a point. Statistics and Probability Letters, 38, 305-309.

50. Hazelton, M.L. (1998). Some remarks on Stochastic User Equilibrium. Trans-portation Research Part B, 32, 101-108.

51. Hazelton, M.L. (1996). Bandwidth selection for local density estimators. Scandi-navian Journal of Statistics, 23, 221-232.

52. Hazelton, M.L. (1996). Optimal rates for local bandwidth selection. Journal ofNonparametric Statistics, 7, 57-66.

53. Hazelton, M.L. (1995). Improved Monte Carlo inference for models with additiveerror. Statistics and Computing, 5, 343-350.

ProposalStandard


InitialsTM


PanelMIS




Professor Adrian John BaddeleyPresent position Professor of Computational StatisticsOrganisation/Employer Curtin UniversityContact address School of Mathematics & Statistics

GPO Box U1987Perth, Western Australia Post code 6845

Work telephone +61 410 447 821 Mobile +61 410 447 821Email [email protected] website(if applicable)

www.spatstat.org

1b. Academic qualifications

1980, PhD, Mathematical Statistics, Cambridge University1976, BA(Hons), Pure Mathematics and Statistics, Australian National University.

1c. Professional positions held

2015–present, Professor of Computational Statistics, Curtin University.2013–2014, Research Professor, Centre for Exploration Targeting, University of WesternAustralia.2010–2012, Research Scientist, CSIRO Mathematics Informatics & Statistics, Perth, Aus-tralia.1994-2010, Full Professor of Statistics, University of Western Australia.1988–1994, Research group leader, CWI (Centre for Mathematics and Computer Science),Amsterdam, Netherlands.1985-1988, Research scientist, CSIRO Division of Mathematics & Statistics, Sydney, Aus-tralia.1982–1985, Lecturer in Statistics, University of Bath, UK.1979–1982, Research Fellow of Trinity College, Cambridge, UK.

1d. Present research/professional speciality

Statistical methodology for spatial data. Statistical computing and statistical software.

1e. Total years research experience: 42 years


2017, John Curtin Distinguished Professorship, Curtin University.2015, Honorary DSc, Aalborg University.2013, Australian Research Council, Discovery Outstanding Researcher Award2008, Matheron Lecturer, International Association for Mathematical Geology.2004, Pitman Medal, Statistical Society of Australia.2001, Centenary Medal, Australian Government.2001, Hannan Medal, Australian Academy of Science.2000, elected Fellow of the Australian Academy of Science.

ProposalStandard


InitialsTM


PanelMIS


1995, Medal of the Australian Mathematical Society.1991, Adjunct Professor, University of Leiden, Netherlands.1979, Prize Research Fellowship, Trinity College Cambridge.1979, Smith-Knight Prize (1st Class), University of Cambridge.1977, External Research Studentship, Trinity College Cambridge.1977, Commonwealth Postgraduate Research Scholarship.1976, University Medal, Australian National University.

1g. Total number of peerreviewed publicationsand patents

Journalarticles



Patents

87 2 11 18 0

ProposalStandard


InitialsTM


PanelMIS


PART 22a. Research publications and disseminationPeer reviewed journal articles (selected)

1. S. Rakshit, T.M. Davies, M.M. Moradi, G. McSwiggan, G. Nair, J. Mateu, andA. Baddeley. Fast kernel smoothing of point patterns on a large network us-ing 2D convolution. International Statistical Review, 2019. In press. DOI:

10.1111/insr.12327.2. S. Rakshit, A. Baddeley, and G. Nair. Efficient code for second-order analysis of

events on a linear network. Journal of Statistical Software, 2019. In press.3. K. Hingee, A. Baddeley, P. Caccetta and G. Nair. Computation of lacunarity from

covariance of spatial binary maps. Journal of Agricultural, Biological and Environ-mental Statistics 24: 264-288. 2019.

4. M. Moradi, O. Cronie, E. Rubak, R. Lachieze-Rey, J. Mateu and A. Baddeley.Resample-smoothing of Voronoi intensity estimators. Statistics and Computing2019, In press. Published online 22 january 2019.

5. A. Baddeley, E. Rubak and R. Turner. Leverage and influence diagnostics forGibbs spatial point processes. Spatial Statistics 29: 15–48, 2019.

6. T.M. Davies and A. Baddeley. Fast computation of spatially adaptive kernel esti-mates. Statistics and Computing 28: 937-956, 2018.

7. A. Baddeley. Local composite likelihood for spatial point processes. Spatial Statis-tics, 22:261–295, 2017.

8. A. Baddeley, A. Hardegen, T. Lawrence, R.K. Milne, G. Nair, and S. Rakshit. Ontwo-stage Monte Carlo tests of composite hypotheses. Computational Statisticsand Data Analysis, 114:75–87, 2017.

9. A. Baddeley and G. Nair. Poisson-saddlepoint approximation for Gibbs point pro-cesses with infinite-order interaction: in memory of Peter Hall. Journal of AppliedProbability, 54(4):1008–1026, December 2017.

10. S. Rakshit, G. Nair, and A. Baddeley. Second-order analysis of point patterns ona network using any distance metric. Spatial Statistics, 22(1):129–154, 2017.

11. A. Baddeley, G. Nair, S. Rakshit, and G. McSwiggan. ‘Stationary’ point processesare uncommon on linear networks. STAT, 6(1):68–78, 2017.

12. A. Baddeley, R. Turner, and E. Rubak. Adjusted composite likelihood ratio test forspatial Gibbs point processes. Journal of Statistical Computation and Simulation,86(5):922–941, 2016.

13. T. Lawrence, A. Baddeley, R.K. Milne, and G. Nair. Point pattern analysis on aregion of a sphere. Stat, 5(1):144–157, 2016.

14. G. McSwiggan, A. Baddeley, and G. Nair. Kernel density estimation on a linearnetwork. Scandinavian Journal of Statistics, 44(2):324–345, 2016.

15. I.W. Renner, J. Elith, A. Baddeley, W. Fithian, T. Hastie, S.J. Phillips, G. Popovic,and D.I. Warton. Point process models for presence-only analysis. Methods inEcology and Evolution, 6(4):366–379, 2015.

16. A. Baddeley, J.-F. Coeurjolly, E. Rubak, and R. Waagepetersen. Logistic regres-sion for spatial Gibbs point processes. Biometrika, 101(2):377–392, 2014.

17. A. Baddeley, P.J. Diggle, A. Hardegen, T. Lawrence, R.K. Milne, and G. Nair. Ontests of spatial pattern based on simulation envelopes. Ecological Monographs,84(3):477–489, 2014.

18. A. Baddeley, A. Jammalamadaka, and G. Nair. Multitype point process analysisof spines on the dendrite network of a neuron. Applied Statistics (Journal of theRoyal Statistical Society, Series C), 63(5):673–694, 2014.

ProposalStandard


InitialsTM


PanelMIS


19. A. Baddeley and R. Turner. Bias correction for parameter estimates of spatial pointprocess models. Journal of Statistical Computation and Simulation, 84:1621–1643,2014.

20. R.S. Anderssen, A. Baddeley, F.R. DeHoog, and G.M. Nair. Solution of an integralequation arising in spatial point process theory. Journal of Integral Equations andApplications, 26(4):437–453, 2014.

21. A. Baddeley, Y.-M. Chang, Y. Song, and R. Turner. Residual diagnostics for covari-ate effects in spatial point process models. Journal of Computational and GraphicalStatistics, 22:886–905, 2013.

22. A. Baddeley, Y.M. Chang, and Y. Song. Leverage and influence diagnostics forspatial point processes. Scandinavian Journal of Statistics, 40:86–104, 2013.

23. A. Baddeley and D. Dereudre. Variational estimators for the parameters of Gibbspoint process models. Bernoulli, 19:905–930, 2013.

24. A. Baddeley, R. Turner, J. Mateu, and A. Bevan. Hybrids of Gibbs point pro-cess models and their implementation. Journal of Statistical Software, 55(11):1–43, 2013.

25. A. Baddeley, Y.M. Chang, Y. Song, and R. Turner. Nonparametric estimation ofthe dependence of a spatial point process on a spatial covariate. Statistics and itsInterface, 5:221–236, 2012.

26. Q.W. Ang, A. Baddeley, and G. Nair. Geometrically corrected second order anal-ysis of events on a linear network, with applications to ecology and criminology.Scandinavian Journal of Statistics, 39:591–617, 2012.

27. A. Baddeley and G. Nair. Fast approximation of the intensity of Gibbs point pro-cesses. Electronic Journal of Statistics, 6:1155–1169, 2012.

28. A. Baddeley and G. Nair. Approximating the moments of a spatial point process.Stat, 1(1):18–30, 2012.

29. A. Baddeley, E. Rubak, and J. Møller. Score, pseudo-score and residual diagnos-tics for spatial point process models. Statistical Science, 26:613–646, 2011.

30. A. Baddeley, M. Berman, N.I. Fisher, A. Hardegen, R.K. Milne, D. Schuhmacher,and R. Turner. Spatial logistic regression and change-of-support for Poisson pointprocesses. Electronic Journal of Statistics, 4:1151–1201, 2010.

31. S.S. Singh, B. Vo, A. Baddeley, and S. Zuyev. Filters for spatial point processes.SIAM Journal on Control and Optimization, 48(4):2275–2295, 2009.

32. A. Baddeley, J. Møller, and A.G. Pakes. Properties of residuals for spatial pointprocesses. Annals of the Institute of Statistical Mathematics, 60:627–649, 2008.

33. J.F. Wallace, M. Canci, X. Wu, and A. Baddeley. Monitoring native vegetation onan urban groundwater supply mound using airborne digital imagery. Journal ofSpatial Science, 53:63–73, 2008. ISSN 1449-8596.

34. A. Baddeley, R. Turner, J. Møller, and M. Hazelton. Residual analysis for spatialpoint processes (with discussion). Journal of the Royal Statistical Society, SeriesB, 67(5):617–666, 2005.

35. R. Foxall and A. Baddeley. Nonparametric measures of association betweena spatial point process and a random set, with geological applications. AppliedStatistics, 51(2):165–182, 2002.

36. A. Baddeley, J. Møller, and R. Waagepetersen. Non- and semiparametric es-timation of interaction in inhomogeneous point patterns. Statistica Neerlandica,54(3):329–350, 2000.

ProposalStandard


InitialsTM


PanelMIS


37. A. Baddeley and R. Turner. Practical maximum pseudolikelihood for spatial pointpatterns (with discussion). Australian and New Zealand Journal of Statistics,42(3):283–322, 2000.

38. Y.C. Chin and A.J. Baddeley. Markov interacting component processes. Advancesin Applied Probability, 32:597–619, 2000.

39. K. Schladitz and A.J. Baddeley. A third order point process characteristic. Scan-dinavian Journal of Statistics, 27:657–671, 2000.

40. Y.C. Chin and A.J. Baddeley. On connected component Markov point processes.Advances in Applied Probability, 31:279–282, 1999.

41. W.S. Kendall, M.N.M. van Lieshout, and A.J. Baddeley. Quermass-interaction pro-cesses: conditions for stability. Advances in Applied Probability, 31:315–342, 1999.

42. M.N.M. van Lieshout and A.J. Baddeley. Indices of dependence between types inmultivariate point patterns. Scandinavian Journal of Statistics, 26:511–532, 1999.

43. A.J. Baddeley and R.D. Gill. Kaplan-Meier estimators of interpoint distance distri-butions for spatial point processes. Annals of Statistics, 25:263–292, 1997.

44. A.J. Baddeley, M.N.M. van Lieshout, and J. Møller. Markov properties of clusterprocesses. Advances in Applied Probability, 28:346–355, 1996.

45. A.J. Baddeley and M.N.M. van Lieshout. Area-interaction point processes. Annalsof the Institute of Statistical Mathematics, 47:601–619, 1995.

46. M.N.M. van Lieshout and A.J. Baddeley. A nonparametric measure of spatial in-teraction in point patterns. Statistica Neerlandica, 50:344–361, 1996.

47. A.J. Baddeley, R.A. Moyeed, C.V. Howard, and A. Boyde. Analysis of a three-dimensional point pattern with replication. Applied Statistics, 42(4):641–668, 1993.

48. R.A. Moyeed and A.J. Baddeley. Stochastic approximation of the MLE for a spatialpoint pattern. Scandinavian Journal of Statistics, 18:39–50, 1991.

49. A.J. Baddeley and J. Møller. Nearest-neighbour Markov point processes and ran-dom sets. International Statistical Review, 57:89–121, 1989.

50. A.J. Baddeley and B.W. Silverman. A cautionary example on the use of second-order methods for analyzing point patterns. Biometrics, 40:1089–1094, 1984.

Peer reviewed books1. A. Baddeley, E. Rubak, and R. Turner. Spatial Point Patterns: Methodology and

Applications with R. Chapman and Hall/CRC, London, 2015.2. A. Baddeley and E.B. Vedel Jensen. Stereology for Statisticians. Chapman and

Hall/CRC, London, 2005.Peer reviewed book chapters, books edited (selected)

1. A. Baddeley. Modelling strategies. In A.E. Gelfand, P.J. Diggle, M. Fuentes, andP. Guttorp, editors, Handbook of Spatial Statistics, chapter 20, pages 339–369.CRC Press, Boca Raton, 2010.

2. A. Baddeley. Multivariate and marked point processes. In A.E. Gelfand, P.J. Dig-gle, M. Fuentes, and P. Guttorp, editors, Handbook of Spatial Statistics, chapter 21,pages 371–402. CRC Press, Boca Raton, 2010.

3. A. Baddeley. Spatial point processes and their applications. In A. Baddeley,I. Barany, R. Schneider, and W. Weil, editors, Stochastic Geometry: Lecturesgiven at the C.I.M.E. Summer School held in Martina Franca, Italy, September 13–18, 2004, Lecture Notes in Mathematics 1892 (subseries: Fondazione C.I.M.E.,Firenze), pages 1–75. Springer-Verlag, 2006.

4. M.N.M. van Lieshout and A.J. Baddeley. Extrapolating and interpolating spatialpatterns. In A.B. Lawson and D.G.T. Denison, editors, Spatial cluster modelling,chapter 4, pages 61–86. Chapman and Hall/CRC Press, Boca Raton, 2002.

ProposalStandard


InitialsTM


PanelMIS


Documents

Proposal Contact PI's Surname Initials Application …MARSDEN FUND FULL RESEARCH PROPOSAL Standard Application Form A new generation of statistical models for spatial point process